TECHNICALNOTESEVEN
technicalnote
Assignable variation defined 3 0 0
Common variation defined
Variation around Us 3 0 1
Upper and lower specification or tolerance limits defined
Process Capability 3 0 2
Capability index (Cpk) Capability index (Cpk) defined
Process Control Procedures 3 0 5
Process control with attribute Statistical process control (SPC) defined
measurements: using p charts Attributes defined
Process control with variable measurements: Variables defined
using X
­
and R charts
How to construct X
­
and R charts
Acceptance Sampling 3 1 1
Design of a single sampling plan for attributes
Operating characteristic curves
Conclusion 3 1 4
technical note seven
PROCESS CAPABILITY AND
STATISTICAL QUALITY
CONTROL
cha06369_tn07.qxd 2/12/03 7:01 PM Page 299
This technical note on statistical quality control (SQC) covers the quantitative aspects of
quality management. In general, SQC is a number of different techniques designed to
evaluate quality from a conformance view. That is, how well are we doing at meeting the
specifications that have been set during the design of the parts or services that we are providing?
Managing quality performance using SQC techniques usually involves periodic
sampling of a process and analysis of these data using statistically derived performance
criteria.
As you will see, SQC can be applied to both manufacturing and service processes. Here
are some examples of the types of situations where SQC can be applied:
* How many paint defects are there in the finish of a car? Have we improved our painting
process by installating a new sprayer?
* How long does it take to execute market orders in our Web-based trading system?
Has the installation of a new server improved the service? Does the performance of
the system vary over the trading day?
* How well are we able to maintain the dimensional tolerance on our three-inch ball
bearing assembly? Given the variability of our process for making this ball bearing,
how many defects would we expect to produce per million bearings that we make?
* How long does it take for customers to be served from our drive-through window
during the busy lunch period?
Processes that provide goods and services usually exhibit some variation in their output.
This variation can be caused by many factors, some of which we can control and others that
are inherent in the process. Variation that is caused by factors that can be clearly identified
and possibly even managed is called assignable variation. For example, variation caused by
workers not being equally trained or by improper machine adjustment is assignable variation.
Variation that is inherent in the process itself is called common variation. Common
variation is often referred to as random variation and may be the result of the type of equipment
used to complete a process, for example.
As the title of this technical note implies, this material requires an understanding of very
basic statistics. Recall from your study of statistics involving numbers that are normally
distributed the definition of the mean and standard deviation. The mean is just the average
value of a set of numbers. Mathematically this is
[TN7.1] X =
N
i=1
xi /N
300 section 2 PRODUCT DESIGN AND PROCESS SELECTION
Servi
ce
IN MONITORING A PROCESS USING SQC,
WORKERS TAKE A SAMPLE WHERE THE DIAMETERS
ARE MEASURED AND THE SAMPLE MEAN IS
CALCULATED AND PLOTTED. INVESTMENTS IN
MACHINERY, TECHNOLOGY, AND EDUCATION ARE
DESIGNED TO REDUCE THE NUMBER OF DEFECTS
THAT THE PROCESS PRODUCES.
Vol. I
"Quality"
Vol. VII
"Manufacturing Quality at
Honda"
"A Day in the Life of Quality
at Honda"
"SPC at Honda"
Assignable variation
Common variation
cha06369_tn07.qxd 2/12/03 7:01 PM Page 300
PROCESS CAPABILITY AND STATISTICAL QUALITY CONTROL technical note 301
where:
xi = Observed value
N = Total number of observed values
The standard deviation is
[TN7.2]  =
N
i=1
(xi - X)2
N
In monitoring a process using SQC, samples of the process output would be taken, and
sample statistics calculated. The distribution associated with the samples should exhibit the
same kind of variability as the actual distribution of the process, although the actual variance
of the sampling distribution would be less. This is good because it allows the quick
detection of changes in the actual distribution of the process. The purpose of sampling is
to find when the process has changed in some nonrandom way, so that the reason for the
change can be quickly determined.
In SQC terminology, sigma is often used to refer to the sample standard deviation. As
you will see in the examples, sigma is calculated in a few different ways, depending on the
underlying theoretical distribution (i.e., a normal distribution or a Poisson distribution).
VA R I AT I O N A R O U N D U S
G G G It is generally accepted that as variation is reduced, quality is improved. Sometimes
that knowledge is intuitive. If a train is always on time, schedules can be planned
more precisely. If clothing sizes are consistent, time can be saved by ordering from a
catalog. But rarely are such things thought about in terms of the value of low variability.
With engineers, the knowledge is better defined. Pistons must fit cylinders, doors must fit
openings, electrical components must be compatible, and boxes of cereal must have the
right amount of raisins--otherwise quality will be unacceptable and customers will be
dissatisfied.
However, engineers also know that it is impossible to have zero variability. For this reason,
designers establish specifications that define not only the target value of something but
also acceptable limits about the target. For example, if the aim value of a dimension is
10 inches, the design specifications might then be 10.00 inches 0.02 inch. This would tell
the manufacturing department that, while it should aim for exactly 10 inches, anything between
9.98 and 10.02 inches is OK. These design limits are often referred to as the upper
and lower specification limits or the upper and lower tolerance limits.
A traditional way of interpreting such a specification is that any part that falls within the
allowed range is equally good, whereas any part falling outside the range is totally bad.
This is illustrated in Exhibit TN7.1. (Note that the cost is zero over the entire specification
range, and then there is a quantum leap in cost once the limit is violated.)
Genichi Taguchi, a noted quality expert from Japan, has pointed out that the traditional
view illustrated in Exhibit TN7.1 is nonsense for two reasons:
1 From the customer's view, there is often practically no difference between a product
just inside specifications and a product just outside. Conversely, there is a far greater
difference in the quality of a product that is the target and the quality of one that is
near a limit.
2 As customers get more demanding, there is pressure to reduce variability. However,
Exhibit TN7.1 does not reflect this logic.
Taguchi suggests that a more correct picture of the loss is shown in Exhibit TN7.2.
Notice that in this graph the cost is represented by a smooth curve. There are dozens of
Upper and lower specification
or tolerance limits
cha06369_tn07.qxd 2/12/03 7:01 PM Page 301
302 section 2 PRODUCT DESIGN AND PROCESS SELECTION
illustrations of this notion: the meshing of gears in a transmission, the speed of photographic
film, the temperature in a workplace or department store. In nearly anything that
can be measured, the customer sees not a sharp line, but a gradation of acceptability away
from the "Aim" specification. Customers see the loss function as Exhibit TN7.2 rather than
Exhibit TN7.1.
Of course, if products are consistently scrapped when they are outside specifications,
the loss curve flattens out in most cases at a value equivalent to scrap cost in the ranges
outside specifications. This is because such products, theoretically at least, will never be
sold so there is no external cost to society. However, in many practical situations, either the
process is capable of producing a very high percentage of product within specifications
and 100 percent checking is not done, or if the process is not capable of producing within
specifications, 100 percent checking is done and out-of-spec products can be reworked to
bring them within specs. In any of these situations, the parabolic loss function is usually a
reasonable assumption.
P R O C E S S C A PA B I L I T Y
G G G Taguchi argues that being within tolerance is not a yes/no decision, but rather a
continuous function. The Motorola quality experts, on the other hand, argue that the process
used to produce a good or deliver a service should be so good that the probability of generating
a defect should be very, very low. Motorola made process capability and product
design famous by adopting six-sigma limits. When we design a part, we specify that certain
dimensions should be within the upper and lower tolerance limits.
As a simple example, assume that we are designing a bearing for a rotating shaft--say
an axle for the wheel of a car. There are many variables involved for both the bearing and
the axle--for example, the width of the bearing, the size of the rollers, the size of the axle,
the length of the axle, how it is supported, and so on. The designer specifies tolerances for
each of these variables to ensure that the parts will fit properly. Suppose that initially a
E X H I B I T TN7.1
A Traditional View of the Cost
of Variability
High
Zero
Incremental
cost to
society of
variability
Lower
spec
Upper
spec
Aim
spec
E X H I B I T TN7.2
Taguchi's View of the Cost of
Variability Incremental
cost to
society of
variability
Lower
spec
Upper
spec
Aim
spec
Zero
High
cha06369_tn07.qxd 2/12/03 7:01 PM Page 302
PROCESS CAPABILITY AND STATISTICAL QUALITY CONTROL technical note 303
design is selected and the diameter of the bearing is set at 1.250 inches
0.005 inch. This means that acceptable parts may have a diameter that
varies between 1.245 and 1.255 inches (which are the lower and upper
tolerance limits).
Next, consider the process in which the bearing will be made. Leťs say
that by running some tests, we determine the machine output to have a
standard deviation or sigma equal to 0.002 inch. What this means is that
our process does not make each bearing exactly the same size.Assume that
we are monitoring the process such that any bearings that are more than
three standard deviations (0.006 inch) above or below 1.250 inches are
rejected. This means that we will produce parts that vary between 1.244
and 1.256 inches.As we can see, our process limits are greater than the tolerance
limits specified by our designer. This is not good, because we will
produce some parts that do not meet specifications.
Motorola insists that a process making a part must be capable of operating
so that the design tolerances are six standard deviations away from the
process mean. For our bearing, this would mean that our process variation
would need to be less than or equal to 0.00083 inch (remember our tolerance
was 0.005, which, when divided by 6, is 0.00083). To reduce the
variation in the process, we would need to find some better method for controlling
the formation of the bearing. Of course, another option would be to
redesign the axle assembly so that such perfect bearings are not needed.
We can show the six-sigma limits using an exhibit.Assume that we have
changed the process to produce with 0.00083 variation. Now, the design
limits and the process limits are acceptable according to Motorola standards.
Leťs assume that the bearing diameter follows a bell-shaped normal distribution as
in Exhibit TN7.3. From our knowledge of the normal distribution, we know that 99.7 percent
of the bell-shaped curve falls within 3 sigma. We would expect only about three
parts in 1,000 to fall outside of the three-sigma limits. The tolerance limits are another three
sigma out from these control limits! In this case, the actual number of parts we would
expect to produce outside the tolerance limits is only two parts per billion!
Suppose the central value of the process output shifts away from the mean. ExhibitTN7.4
shows the mean shifted one standard deviation closer to the upper specification limit. This
E X H I B I T TN7.3
Process Capability
Standard deviation ( )
­6
Lower
tolerance
limit
­5 ­4 ­3 ­2 ­1 0 1 2 3 4 5 6
Upper
tolerance
limit
E X H I B I T TN7.4
Process Capability with a Shift
in the Process Mean
­6
Lower
tolerance
limit
­5 ­4 ­3 ­2 ­1 0 1 2 3 4 5 6
Upper
tolerance
limit
cha06369_tn07.qxd 2/12/03 7:01 PM Page 303
304 section 2 PRODUCT DESIGN AND PROCESS SELECTION
causes a slightly higher number of expected defects, about four parts per million. This is still
pretty good, by most people's standards. We use a calculation called the capability index to
measure how well our process is capable of producing relative to the design tolerances. We
describe how to calculate this index in the next section.
C A P A B I L I T Y I N D E X (Cpk)
The capability index (Cpk) shows how well the parts being produced fit into the range specified
by the design limits. If the design limits are larger than the three sigma allowed in the
process, then the mean of the process can be allowed to drift off-center before readjustment,
and a high percentage of good parts will still be produced.
Referring to Exhibits TN7.3 and TN7.4, the capability index (Cpk) is the position of the
mean and tails of the process relative to design specifications. The more off-center, the
greater the chance to produce defective parts.
Because the process mean can shift in either direction, the direction of shift and its distance
from the design specification set the limit on the process capability. The direction of
shift is toward the smaller number.
Formally stated, the capability index (Cpk) is calculated as the smaller number as follows:
[TN7.3] Cpk = min
X - LTL
3
or
UTL - X
3
For simplicity, leťs assume our process mean is one inch and  = .001 ( is the symbol for
standard deviation). Further, the process mean is exactly in the center as in Exhibit TN7.3.
Then for X = 1.000
Cpk = min
1.000 - .994
3(.001)
or
1.006 - 1.000
3(.001)
=
= min
.006
.003
= 2 or
.006
.003
= 2
Because the mean is in the center, the two calculations are the same and equal to 2. If the
mean shifted to +1.5 or 1.0015, then for X = 1.0015
Cpk = min
1.0015 - .994
3(.001)
or
1.006 - 1.0015
3(.001)
=
= min [2.5 or 1.5]
Cpk = 1.5, which is the smaller number.
This tells us that the process mean has shifted to the right similar to Exhibit TN7.4, but
parts are still well within design limits.
Assuming that the process is producing with a consistent standard deviation and the
process is centered exactly between the design limits, as in Exhibit TN7.3, Birch calculated
the fraction of defective units that would fall outside various design limits as follows:1
DESIGN LIMITS DEFECTIVE PARTS FRACTION DEFECTIVE
1 317 per thousand .3173
2 45 per thousand .0455
3 2.7 per thousand .0027
4 63 per million .000063
5 574 per billion .000000574
6 2 per billion .000000002
Capability index (Cpk)
cha06369_tn07.qxd 2/12/03 7:01 PM Page 304
PROCESS CAPABILITY AND STATISTICAL QUALITY CONTROL technical note 305
G G G Process control is concerned with monitoring quality while the product or service
is being produced. Typical objectives of process control plans are to provide timely
information on whether currently produced items are meeting design specifications and
to detect shifts in the process that signal that future products may not meet specifications.
Statistical process control (SPC) involves testing a random sample of output from a
process to determine whether the process is producing items within a preselected range.
The examples given so far have all been based on quality characteristics (or variables)
that are measurable, such as the diameter or weight of a part. Attributes are quality characteristics
that are classified as either conforming or not conforming to specification.
Goods or services may be observed to be either good or bad, or functioning or malfunctioning.
For example, a lawnmower either runs or it doesn't; it attains a certain level of
torque and horsepower or it doesn't. This type of measurement is known as sampling by
attributes. Alternatively, a lawnmower's torque and horsepower can be measured as an
amount of deviation from a set standard. This type of measurement is known as sampling
by variables. The following section describes some standard approaches to controlling
processes: first an approach useful for attribute measures and then an approach for
variable measures. Both of these techniques result in the construction of control charts.
Exhibit TN7.5 shows some examples for how control charts can be analyzed to understand
how a process is operating.
P R O C E S S C O N T R O L W I T H AT T R I B U T E
M E A S U R E M E N T S : U S I N G p C H A R T S
Measurement by attributes means taking samples and using a single decision--the item is
good, or it is bad. Because it is a yes or no decision, we can use simple statistics to create a
p chart with an upper control limit (UCL) and a lower control limit (LCL). We can draw
these control limits on a graph and then plot the fraction defective of each individual sample
tested. The process is assumed to be working correctly when the samples, which are
taken periodically during the day, continue to stay between the control limits.
[TN7.4] p =
Total number of defects from all samples
Number of samples × Sample size
[TN7.5] sp =
p(1 - p)
n
[TN7.6] UCL = p + zsp
[TN7.7] LCL = p - zsp
where p is the fraction defective, sp is the standard deviation, n is the sample size, and z is
the number of standard deviations for a specific confidence. Typically, z = 3 (99.7 percent
confidence) or z = 2.58 (99 percent confidence) is used.
S i z e o f t h e S a m p l e The size of the sample must be large enough to allow counting
of the attribute. For example, if we know that a machine produces 1 percent defects,
then a sample size of five would seldom capture a defect. A rule of thumb when setting up
a p chart is to make the sample large enough to expect to count the attribute twice in each
Attributes
P R O C E S S C O N T R O L P R O C E D U R E S
Statistical process control (SPC)
Motorola's design limit of six sigma with a shift of the process off the mean by 1.5
(Cpk = 1.5) gives 3.4 defects per million. If the mean is exactly in the center (Cpk = 2), then
2 defects per billion are expected, as the table above shows.
Inter
active Op
erationsM
anageme
nt
cha06369_tn07.qxd 2/12/03 7:01 PM Page 305
306 section 2 PRODUCT DESIGN AND PROCESS SELECTION
sample. So an appropriate sample size if the defect rate were approximately 1 percent would
be 200 units. One final note: In the calculations shown in equations TN7.4­7.7 the assumption
is that the sample size is fixed. The calculation of the standard deviation depends on this
assumption. If the sample size varies, the standard deviation and upper and lower control
limits should be recalculated for each sample.
EXAMPLE TN7.1: Control Chart Design
An insurance company wants to design a control chart to monitor whether insurance claim forms are
being completed correctly. The company intends to use the chart to see if improvements in the design
of the form are effective. To start the process the company collected data on the number of incorrectly
completed claim forms over the past 10 days. The insurance company processes thousands of these
forms each day, and due to the high cost of inspecting each form, only a small representative sample
was collected each day. The data and analysis are shown in Exhibit TN7.6.
Servi
ce
E X H I B I T TN7.5 Control Chart Evidence for Investigation
Upper control limit
Central line
Lower control limit
Normal behavior. One plot out above.
Investigate for cause of poor
performance.
One plot out below.
Investigate for cause of
the low value.
Upper control limit
Central line
Lower control limit
Two plots near upper control.
Investigate for cause of poor
performance.
Two plots near lower
control. Investigate for
cause.
Run of five above central line.
Investigate for cause of
sustained poor performance.
Upper control limit
Central line
Lower control limit
Run of five below central line.
Investigate for cause of
sustained poor performance.
Trend in either direction five
plots. Investigate for cause of
progressive change.
Erratic behavior.
Investigate.
Upper control limit
Central line
Lower control limit
Sudden change in level.
Investigate for cause.
Time Time Time
cha06369_tn07.qxd 2/12/03 7:01 PM Page 306
PROCESS CAPABILITY AND STATISTICAL QUALITY CONTROL technical note 307
SOLUTION
To construct the control chart, first calculate the overall fraction defective from all samples. This sets
the centerline for the control chart.
p =
Total number of defects from all samples
Number of samples × Sample size
=
91
3000
= .03033
Next calculate the sample standard deviation:
sp =
p(1 - p )
n
=
.03033(1 - .03033)
300
= .00990
Finally, calculate the upper and lower control limits. A z-value of 3 gives 99.7 percent confidence that
the process is within these limits.
UCL = p + 3sp = .03033 + 3(.00990) = .06004
LCL = p - 3sp = .03033 - 3(.00990) = .00063
The calculations in Exhibit TN7.6, including the control chart, are included in the spreadsheet
SPC.xls.
ˇ
P R O C E S S C O N T R O L W I T H V A R I A B L E
M E A S U R E M E N T S : U S I N G
­
X A N D R C H A R T S
X and R (range) charts are widely used in statistical process control.
In attribute sampling, we determine whether something is good or bad, fits or doesn't
fit--it is a go/no-go situation. In variables sampling, however, we measure the actual
weight, volume, number of inches, or other variable measurements, and we develop control
charts to determine the acceptability or rejection of the process based on those measurements.
For example, in attribute sampling we might decide that if something is over
10 pounds we will reject it and under 10 pounds we will accept it. In variable sampling we
E
xcel: SPC.
xls
Variables
E X H I B I T TN7.6
NUMBER OF
FORMS
NUMBER COMPLETED FRACTION
SAMPLE INSPECTED INCORRECTLY DEFECTIVE
1 300 10 0.03333
2 300 8 0.02667
3 300 9 0.03000
4 300 13 0.04333
5 300 7 0.02333
6 300 7 0.02333
7 300 6 0.02000
8 300 11 0.03667
9 300 12 0.04000
10 300 8 0.02667
Totals 3000 91 0.03033
Sample standard deviation 0.00990
Insurance Company Claim Form
1 2 3 4 5 6 7 8 9 10
0.00000
0.00500
0.01000
0.01500
0.02000
0.02500
0.03000
0.03500
0.04000
0.04500
0.05000
0.05500
0.06000
0.06500
Sample
Upper control limit
Lower control limit
cha06369_tn07.qxd 2/12/03 7:01 PM Page 307
308 section 2 PRODUCT DESIGN AND PROCESS SELECTION
measure a sample and may record weights of 9.8 pounds or 10.2 pounds. These values are
used to create or modify control charts and to see whether they fall within the acceptable
limits.
There are four main issues to address in creating a control chart: the size of the samples,
number of samples, frequency of samples, and control limits.
S i z e o f S a m p l e s For industrial applications in process control involving the measurement
of variables, it is preferable to keep the sample size small. There are two main
reasons. First, the sample needs to be taken within a reasonable length of time; otherwise,
the process might change while the samples are taken. Second, the larger the sample, the
more it costs to take.
Sample sizes of four or five units seem to be the preferred numbers. The means of samples
of this size have an approximately normal distribution, no matter what the distribution
of the parent population looks like. Sample sizes greater than five give narrower control
limits and thus more sensitivity. For detecting finer variations of a process, it may be necessary,
in fact, to use larger sample sizes. However, when sample sizes exceed 15 or so, it
would be better to use X charts with standard deviation  rather than X charts with the
range R as we use in Example TN7.2.
N u m b e r o f S a m p l e s Once the chart has been set up, each sample taken can be
compared to the chart and a decision can be made about whether the process is acceptable.
To set up the charts, however, prudence and statistics suggest that 25 or so samples
be taken.
F r e q u e n c y o f S a m p l e s How often to take a sample is a trade-off between the
cost of sampling (along with the cost of the unit if it is destroyed as part of the test) and
the benefit of adjusting the system. Usually, it is best to start off with frequent sampling
of a process and taper off as confidence in the process builds. For example, one might
start with a sample of five units every half hour and end up feeling that one sample per
day is adequate.
C o n t r o l L i m i t s Standard practice in statistical process control for variables is to
set control limits three standard deviations above the mean and three standard deviations
below. This means that 99.7 percent of the sample means are expected to fall within these
control limits (that is, within a 99.7 percent confidence interval). Thus, if one sample
mean falls outside this obviously wide band, we have strong evidence that the process is
out of control.
PROCESS CONTROL CHARTS CAN BE GENERATED
WITH COMPUTERS OR MANUALLY. SAMPLES ARE
TAKEN FROM THE PROCESS AT STATED TIME
INTERVALS AND THEIR AVERAGE PARAMETER
VALUES ARE PLOTTED ON THE CHARTS. AS LONG
AS THESE VALUES REMAIN WITHIN ESTABLISHED
CONTROL LIMITS, THE PROCESS IS IN CONTROL.
cha06369_tn07.qxd 2/12/03 7:01 PM Page 308
PROCESS CAPABILITY AND STATISTICAL QUALITY CONTROL technical note 309
H O W T O C O N S T R U C T
­
X A N D R C H A R T S
If the standard deviation of the process distribution is known, the X chart may be defined:
[TN7.8] UCLX = X + zsX and LCLX = X - zsX
where
sX = s/

n = Standard deviation of sample means
s = Standard deviation of the process distribution
n = Sample size
X = Average of sample means or a target value set for the process
z = Number of standard deviations for a specific confidence level (typically, z = 3)
An X chart is simply a plot of the means of the samples that were taken from a process. X
is the average of the means.
In practice, the standard deviation of the process is not known. For this reason, an approach
that uses actual sample data is commonly used. This practical approach is described
in the next section.
An R chart is a plot of the range within each sample. The range is the difference between
the highest and the lowest numbers in that sample. R values provide an easily calculated
measure of variation used like a standard deviation. An R chart is the average of the range
of each sample. More specifically defined, these are
[Same as TN7.1] X =
n
i=1
Xi
n
where
X = Mean of the sample
i = Item number
n = Total number of items in the sample
[TN7.9] X =
m
j=1
Xj
m
where
X = The average of the means of the samples
j = Sample number
m = Total number of samples
Rj = Difference between the highest and lowest measurement in the sample
R = Average of the measurement differences R for all samples, or
[TN7.10] R =
m
j=1
Rj
m
E. L. Grant and R. Leavenworth computed a table (Exhibit TN7.7) that allows us to easily
compute the upper and lower control limits for both the X chart and the R chart.2
These
are defined as
[TN7.11] Upper control limit for X = X + A2 R
[TN7.12] Lower control limit for X = X - A2 R
[TN7.13] Upper control limit for R = D4 R
[TN7.14] Lower control limit for R = D3 R
cha06369_tn07.qxd 2/12/03 7:01 PM Page 309
310 section 2 PRODUCT DESIGN AND PROCESS SELECTION
E X H I B I T TN7.7
Factor for Determining from R
­
the Three-Sigma Control Limits
for X
­
and R Charts
FACTORS FOR R CHART
NUMBER OF
OBSERVATIONS FACTOR FOR LOWER CONTROL UPPER CONTROL
IN SUBGROUP X
­
CHART LIMIT LIMIT
n A2 D3 D4
2 1.88 0 3.27
3 1.02 0 2.57
4 0.73 0 2.28
5 0.58 0 2.11
6 0.48 0 2.00
7 0.42 0.08 1.92
8 0.37 0.14 1.86
9 0.34 0.18 1.82
10 0.31 0.22 1.78
11 0.29 0.26 1.74
12 0.27 0.28 1.72
13 0.25 0.31 1.69
14 0.24 0.33 1.67
15 0.22 0.35 1.65
16 0.21 0.36 1.64
17 0.20 0.38 1.62
18 0.19 0.39 1.61
19 0.19 0.40 1.60
20 0.18 0.41 1.59
Upper control limit for X
­
= UCLX
­ X
­­
+ A2R
­
Lower control limit for X
­
= LCLX
­ X
­­
- A2R
­
Upper control limit for R = UCLR D4R
­
Lower control limit for R = LCLR D3R
­
Note: All factors are based on the normal distribution.
E X H I B I T TN7.8
Measurements in Samples of
Five from a Process
SAMPLE NUMBER EACH UNIT IN SAMPLE AVERAGE X
­
RANGE R
1 10.60 10.40 10.30 9.90 10.20 10.28 .70
2 9.98 10.25 10.05 10.23 10.33 10.17 .35
3 9.85 9.90 10.20 10.25 10.15 10.07 .40
4 10.20 10.10 10.30 9.90 9.95 10.09 .40
5 10.30 10.20 10.24 10.50 10.30 10.31 .30
6 10.10 10.30 10.20 10.30 9.90 10.16 .40
7 9.98 9.90 10.20 10.40 10.10 10.12 .50
8 10.10 10.30 10.40 10.24 10.30 10.27 .30
9 10.30 10.20 10.60 10.50 10.10 10.34 .50
10 10.30 10.40 10.50 10.10 10.20 10.30 .40
11 9.90 9.50 10.20 10.30 10.35 10.05 .85
12 10.10 10.36 10.50 9.80 9.95 10.14 .70
13 10.20 10.50 10.70 10.10 9.90 10.28 .80
14 10.20 10.60 10.50 10.30 10.40 10.40 .40
15 10.54 10.30 10.40 10.55 10.00 10.36 .55
16 10.20 10.60 10.15 10.00 10.50 10.29 .60
17 10.20 10.40 10.60 10.80 10.10 10.42 .70
18 9.90 9.50 9.90 10.50 10.00 9.96 1.00
19 10.60 10.30 10.50 9.90 9.80 10.22 .80
20 10.60 10.40 10.30 10.40 10.20 10.38 .40
21 9.90 9.60 10.50 10.10 10.60 10.14 1.00
22 9.95 10.20 10.50 10.30 10.20 10.23 .55
23 10.20 9.50 9.60 9.80 10.30 9.88 .80
24 10.30 10.60 10.30 9.90 9.80 10.18 .80
25 9.90 10.30 10.60 9.90 10.10 10.16 .70
X
­­
= 10.21
R
­
= .60
cha06369_tn07.qxd 2/12/03 7:01 PM Page 310
PROCESS CAPABILITY AND STATISTICAL QUALITY CONTROL technical note 311
EXAMPLE TN7.2: X
­
and R Charts
We would like to create X and R charts for a process. Exhibit TN7.8 shows measurements for all
25 samples. The last two columns show the average of the sample X and the range R.
Values for A2, D3, and D4 were obtained from Exhibit TN7.7.
Upper control limit for X = X + A2 R = 10.21 + .58(.60) = 10.56
Lower control limit for X = X - A2 R = 10.21 - .58(.60) = 9.86
Upper control limit for R = D4 R = 2.11(.60) = 1.27
Lower control limit for R = D3 R = 0(.60) = 0
SOLUTION
Exhibit TN7.9 shows the X chart and R chart with a plot of all the sample means and ranges of the
samples. All the points are well within the control limits, although sample 23 is close to the X lower
control limit.
ˇ
A C C E P TA N C E S A M P L I N G
D E S I G N O F A S I N G L E S A M P L I N G P L A N
F O R AT T R I B U T E S
Acceptance sampling is performed on goods that already exist to determine what percentage
of products conform to specifications. These products may be items received from another
company and evaluated by the receiving department, or they may be components that
have passed through a processing step and are evaluated by company personnel either in
production or later in the warehousing function. Whether inspection should be done at all
is addressed in the following example.
Acceptance sampling is executed through a sampling plan. In this section we illustrate
the planning procedures for a single sampling plan--that is, a plan in which the quality is
determined from the evaluation of one sample. (Other plans may be developed using two
or more samples. See J. M. Juran and F. M. Gryna's Quality Planning and Analysis for a
discussion of these plans.)
EXAMPLE TN7.3: Costs to Justify Inspection
Total (100 percent) inspection is justified when the cost of a loss incurred by not inspecting is greater
than the cost of inspection. For example, suppose a faulty item results in a $10 loss and the average
percentage defective of items in the lot is 3 percent.
E X H I B I T TN7.9
UCL 10.55
LCL 9.86
10 12 14 16 18 20 22 24
Sample number
2 4 6 8
10.6
10.5
10.4
10.3
X
=
= 10.2
10.1
10
9.9
9.8
LCL
10 12 14 16 18 20 22 24
Sample number
2 4 6 8
1.4
1.2
1.00
.80
.60
.40
.20
0
UCL 1.26
R
­
X
­
Chart and R Chart
cha06369_tn07.qxd 2/12/03 7:01 PM Page 311
312 section 2 PRODUCT DESIGN AND PROCESS SELECTION
SOLUTION
If the average percentage of defective items in a lot is 3 percent, the expected cost of faulty items is
0.03 × $10, or $0.30 each. Therefore, if the cost of inspecting each item is less than $0.30, the economic
decision is to perform 100 percent inspection. Not all defective items will be removed, however,
because inspectors will pass some bad items and reject some good ones.
The purpose of a sampling plan is to test the lot to either (1) find its quality or (2) ensure that the
quality is what it is supposed to be. Thus, if a quality control supervisor already knows the quality
(such as the 0.03 given in the example), he or she does not sample for defects. Either all of them must
be inspected to remove the defects or none of them should be inspected, and the rejects pass into the
process. The choice simply depends on the cost to inspect and the cost incurred by passing a reject.
ˇ
Asingle sampling plan is defined by n and c, where n is the number of units in the sample
and c is the acceptance number. The size of n may vary from one up to all the items in the lot
(usually denoted as N) from which it is drawn. The acceptance number c denotes the maximum
number of defective items that can be found in the sample before the lot is rejected.
Values for n and c are determined by the interaction of four factors (AQL, , LTPD, and )
that quantify the objectives of the producťs producer and its consumer. The objective of the
producer is to ensure that the sampling plan has a low probability of rejecting good lots. Lots
are defined as high quality if they contain no more than a specified level of defectives, termed
the acceptable quality level (AQL).3
The objective of the consumer is to ensure that the sampling
plan has a low probability of accepting bad lots. Lots are defined as low quality if the
percentage of defectives is greater than a specified amount, termed lot tolerance percent
defective (LTPD). The probability associated with rejecting a high-quality lot is denoted by
the Greek letter alpha () and is termed the producer's risk. The probability associated with
accepting a low-quality lot is denoted by the letter beta () and is termed the consumer's risk.
The selection of particular values forAQL, , LTPD, and  is an economic decision based on
a cost trade-off or, more typically, on company policy or contractual requirements.
There is a humorous story supposedly about Hewlett-Packard during its first dealings
with Japanese vendors, who place great emphasis on high-quality production. HP had
insisted on 2 percent AQL in a purchase of 100 cables. During the purchase agreement,
some heated discussion took place wherein the Japanese vendor did not want this AQL
specification; HP insisted that they would not budge from the 2 percent AQL. The Japanese
vendor finally agreed. Later, when the box arrived, there were two packages inside. One contained
100 good cables. The other package had 2 cables with a note stating: "We have sent
AN EMPLOYEE AT THE STELLENBOSCH WINERY
BOTTLING PLANT CHECKS THE QUALITY OF
NEWLY BOTTLED CHARDONNAY WINE FOR
PROPER VOLUME AND CORK FIT.
cha06369_tn07.qxd 2/12/03 7:01 PM Page 312
PROCESS CAPABILITY AND STATISTICAL QUALITY CONTROL technical note 313
you 100 good cables. Since you insisted on 2 percent AQL, we have enclosed 2 defective
cables in this package, though we do not understand why you want them."
The following example, using an excerpt from a standard acceptance sampling table,
illustrates how the four parameters--AQL, , LTPD, and --are used in developing a sampling
plan.
EXAMPLE TN7.4: Values of n and c
Hi-Tech Industries manufactures Z-Band radar scanners used to detect speed traps. The printed circuit
boards in the scanners are purchased from an outside vendor. The vendor produces the boards to
an AQL of 2 percent defectives and is willing to run a 5 percent risk () of having lots of this level or
fewer defectives rejected. Hi-Tech considers lots of 8 percent or more defectives (LTPD) unacceptable
and wants to ensure that it will accept such poor-quality lots no more than 10 percent of the time
(). A large shipment has just been delivered. What values of n and c should be selected to determine
the quality of this lot?
SOLUTION
The parameters of the problem are AQL = 0.02,  = 0.05, LTPD = 0.08, and  = 0.10. We can use
Exhibit TN7.10 to find c and n.
First, divide LTPD by AQL (0.08 ÷ 0.02 = 4). Then, find the ratio in column 2 that is equal to or
just greater than that amount (4). This value is 4.057, which is associated with c = 4.
Finally, find the value in column 3 that is in the same row as c = 4, and divide that quantity by
AQL to obtain n (1.970 ÷ 0.02 = 98.5).
The appropriate sampling plan is c = 4, n = 99.
ˇ
O P E R A T I N G C H A R A C T E R I S T I C C U R V E S
While a sampling plan such as the one just described meets our requirements for the extreme
values of good and bad quality, we cannot readily determine how well the plan discriminates
between good and bad lots at intermediate values. For this reason, sampling plans
are generally displayed graphically through the use of operating characteristic (OC) curves.
These curves, which are unique for each combination of n and c, simply illustrate the probability
of accepting lots with varying percentages of defectives. The procedure we have followed
in developing the plan, in fact, specifies two points on an OC curve: one point
defined by AQL and 1 - , and the other point defined by LTPD and . Curves for common
values of n and c can be computed or obtained from available tables.4
S h a p i n g t h e O C C u r v e Asampling plan discriminating perfectly between good
and bad lots has an infinite slope (vertical) at the selected value ofAQL. In Exhibit TN7.11,
any percentage defective to the left of 2 percent would always be accepted, and those to the
right, always rejected. However, such a curve is possible only with complete inspection of
all units and thus is not a possibility with a true sampling plan.
An OC curve should be steep in the region of most interest (between the AQL and the
LTPD), which is accomplished by varying n and c. If c remains constant, increasing the sample
size n causes the OC curve to be more vertical. While holding n constant, decreasing c
(the maximum number of defective units) also makes the slope more vertical, moving closer
to the origin.
E X H I B I T TN7.10C LTPD ÷ AQL n  AQL C LTPD ÷ AQL n  AQL
0 44.890 0.052 5 3.549 2.613
1 10.946 0.355 6 3.206 3.286
2 6.509 0.818 7 2.957 3.981
3 4.890 1.366 8 2.768 4.695
4 4.057 1.970 9 2.618 5.426
Excerpt from a Sampling Plan
Table for  = 0.05,  = 0.10
cha06369_tn07.qxd 2/12/03 7:01 PM Page 313
314 section 2 PRODUCT DESIGN AND PROCESS SELECTION
G G G Statistical quality control is a vital topic. Quality has become so important that
statistical quality procedures are expected to be part of successful firms. Sampling plans
and statistical process control are taken as given with the emphasis shifting to broader
aspects (such as eliminating dockside acceptance sampling because of reliable supplier quality,
and employee empowerment transforming much of the process control). World-class
manufacturing companies expect people to understand the basic concepts of the material
presented in this technical note.
C O N C L U S I O N
T h e E f f e c t s o f L o t S i z e The size of the lot that the sample is taken from has relatively
little effect on the quality of protection. Consider, for example, that samples--all of
the same size of 20 units--are taken from different lots ranging from a lot size of 200 units
to a lot size of infinity. If each lot is known to have 5 percent defectives, the probability of
accepting the lot based on the sample of 20 units ranges from about 0.34 to about 0.36. This
means that as long as the lot size is several times the sample size, it makes little difference
how large the lot is. It seems a bit difficult to accept, but statistically (on the average in the
long run) whether we have a carload or box full, we'll get about the same answer. It just
seems that a carload should have a larger sample size. Of course, this assumes that the lot
is randomly chosen and that defects are randomly spread through the lot.
K E Y T E R M S
Assignable variation Deviation in the output of a process that can be
clearly identified and managed.
Common variation Deviation in the output of a process that is random
and inherent in the process itself.
Upper and lower specification or tolerance limits The range of
values in a measure associated with a process that are allowable
given the intended use of the product or service.
E X H I B I T TN7.11
Operating Characteristic Curve
for AQL = 0.02,  = 0.05,
LTPD = 0.08,  = 0.10
Probability of
acceptance
1.0
0.90
0.80
0.70
0.60
0.50
0.40
0.30
0.20
0.10
= 0.05 (producer's risk)
n = 99
c = 4
* = Points specified by sampling plan
ß = 0.10 (consumer's risk)
AQL LTPD
Percentage defective
0 1 2 3 4 5 6 7 8 9 10 11 12 13
cha06369_tn07.qxd 2/12/03 7:01 PM Page 314
PROCESS CAPABILITY AND STATISTICAL QUALITY CONTROL technical note 315
Capability index (Cpk) The ratio of the range of values produced by
a process divided by the range of values allowed by the design
specification.
Statistical process control (SPC) Techniques for testing a random
sample of output from a process to determine whether the process is
producing items within a prescribed range.
Attributes Quality characteristics that are classified as either conforming
or not conforming to specification.
Variables Quality characteristics that are measured in actual weight,
volume, inches, centimeter, or other measure.
F O R M U L A R E V I E W
Mean or average
[TN7.1] X =
N
i=1
xi /N
Standard deviation
[TN7.2]  =
N
i=1
(xi - X)2
N
Capability index
[TN7.3] Cpk = min
X - LTL
3
,
UTL - X
3
Process control charts using attribute measurements
[TN7.4] p =
Total number of defects from all samples
Number of samples × Sample size
[TN7.5] sp =
p(1 - p )
n
[TN7.6] UCL = p + zsp
[TN7.7] LCL = p - zsp
[TN7.8] UCL X = X + zs x and LCL X = X - zsX
Process control X
­
and R charts
[TN7.9] X =
m
j=1
Xj
m
[TN7.10] R =
m
j=1
Rj
m
cha06369_tn07.qxd 2/12/03 7:01 PM Page 315
316 section 2 PRODUCT DESIGN AND PROCESS SELECTION
[TN7.11] Upper control limit for X = X + A2 R
[TN7.12] Lower control limit for X = X - A2 R
[TN7.13] Upper control limit for R = D4 R
[TN7.14] Lower control limit for R = D3 R
S O L V E D P R O B L E M S
SOLVED PROBLEM 1
Completed forms from a particular department of an insurance company were sampled daily to
check the performance quality of that department. To establish a tentative norm for the department,
one sample of 100 units was collected each day for 15 days, with these results:
NUMBER OF NUMBER OF
FORMS WITH FORMS WITH
SAMPLE SAMPLE SIZE ERRORS SAMPLE SAMPLE SIZE ERRORS
1 100 4 9 100 4
2 100 3 10 100 2
3 100 5 11 100 7
4 100 0 12 100 2
5 100 2 13 100 1
6 100 8 14 100 3
7 100 1 15 100 1
8 100 3
a. Develop a p chart using a 95 percent confidence interval (1.96sp).
b. Plot the 15 samples collected.
c. What comments can you make about the process?
Solution
a. p =
46
15(100)
= .0307
sp =
p(1 - p)
n
=
.0307(1 - .0307)
100
=

.0003 = .017
UCL = p + 1.96sp = .031 + 1.96(.017) = .064
LCL = p - 1.96sp = .031 - 1.96(.017) = -.003 or zero
b. The defectives are plotted below.
Proportion
of
defectives
15
Sample number
1413121110987654321
.09
.08
.07
.06
.05
.04
.03
.02
.01
­p
UCL = 0.064
LCL = 0.0
cha06369_tn07.qxd 2/12/03 7:01 PM Page 316
PROCESS CAPABILITY AND STATISTICAL QUALITY CONTROL technical note 317
c. Of the 15 samples, 2 were out of the control limits. Because the control limits were established
as 95 percent, or 1 out of 20, we would say that the process is out of control. It needs to be examined
to find the cause of such widespread variation.
SOLVED PROBLEM 2
Management is trying to decide whether Part A, which is produced with a consistent 3 percent
defective rate, should be inspected. If it is not inspected, the 3 percent defectives will go through
a product assembly phase and have to be replaced later. If all Part A's are inspected, one-third of
the defectives will be found, thus raising the quality to 2 percent defectives.
a. Should the inspection be done if the cost of inspecting is $0.01 per unit and the cost of replacing
a defective in the final assembly is $4.00?
b. Suppose the cost of inspecting is $0.05 per unit rather than $0.01. Would this change your
answer in a?
Solution
Should Part A be inspected?
.03 defective with no inspection.
.02 defective with inspection.
a. This problem can be solved simply by looking at the opportunity for 1 percent improvement.
Benefit = .01($4.00) = $0.04
Cost of inspection = $0.01
Therefore, inspect and save $0.03 per unit.
b. Acost of $0.05 per unit to inspect would be $0.01 greater than the savings, so inspection should
not be performed.
R E V I E W A N D D I S C U S S I O N Q U E S T I O N S
1 The capability index allows for some drifting of the process mean. Discuss what this means in
terms of product quality output.
2 Discuss the purposes of and differences between p charts and X and R charts.
3 In an agreement between a supplier and a customer, the supplier must ensure that all parts are
within tolerance before shipment to the customer. What is the effect on the cost of quality to
the customer?
4 In the situation described in Question 3, what would be the effect on the cost of quality to the
supplier?
5 Discuss the trade-off between achieving a zero AQL (acceptable quality level) and a positive
AQL (such as an AQL of 2 percent).
P R O B L E M S
1 A company currently using an inspection process in its material receiving department is trying
to install an overall cost reduction program. One possible reduction is the elimination
of one inspection position. This position tests material that has a defective content on the
average of 0.04. By inspecting all items, the inspector is able to remove all defects. The
inspector can inspect 50 units per hour. The hourly rate including fringe benefits for this
position is $9. If the inspection position is eliminated, defects will go into product assembly
and will have to be replaced later at a cost of $10 each when they are detected in final
product testing.
a. Should this inspection position be eliminated?
b. What is the cost to inspect each unit?
c. Is there benefit (or loss) from the current inspection process? How much?
2 Ametal fabricator produces connecting rods with an outer diameter that has a 1  .01 inch specification.Amachine
operator takes several sample measurements over time and determines the
sample mean outer diameter to be 1.002 inches with a standard deviation of .003 inch.
a. Calculate the process capability index for this example.
b. What does this figure tell you about the process?
1. a. Not inspecting
cost = $20/hr. Cost
to inspect = $9/hr.
Therefore, inspect.
b. $.18/each.
c. $.22 per unit.
2. a. Cpk = .889.
b. The process is capable but needs
to adjust the mean downward to
achieve the best quality.
cha06369_tn07.qxd 2/12/03 7:01 PM Page 317
318 section 2 PRODUCT DESIGN AND PROCESS SELECTION
3 Ten samples of 15 parts each were taken from an ongoing process to establish a p chart for control.
The samples and the number of defectives in each are shown in the following table:
NUMBER OF NUMBER OF
DEFECTS IN DEFECTS IN
SAMPLE n SAMPLE SAMPLE n SAMPLE
1 15 3 6 15 2
2 15 1 7 15 0
3 15 0 8 15 3
4 15 0 9 15 1
5 15 0 10 15 0
a. Develop a p chart for 95 percent confidence (1.96 standard deviations).
b. Based on the plotted data points, what comments can you make?
4 Output from a process contains 0.02 defective units. Defective units that go undetected into
final assemblies cost $25 each to replace. An inspection process, which would detect and
remove all defectives, can be established to test these units. However, the inspector, who can
test 20 units per hour, is paid $8 per hour, including fringe benefits. Should an inspection
station be established to test all units?
a. What is the cost to inspect each unit?
b. What is the benefit (or loss) from the inspection process?
5 There is a 3 percent error rate at a specific point in a production process. If an inspector is
placed at this point, all the errors can be detected and eliminated. However, the inspector is
paid $8 per hour and can inspect units in the process at the rate of 30 per hour.
If no inspector is used and defects are allowed to pass this point, there is a cost of $10 per
unit to correct the defect later on.
Should an inspector be hired?
6 Resistors for electronic circuits are manufactured on a high-speed automated machine. The
machine is set up to produce a large run of resistors of 1,000 ohms each.
To set up the machine and to create a control chart to be used throughout the run, 15 samples
were taken with four resistors in each sample. The complete list of samples and their measured
values are as follows:
SAMPLE NUMBER READINGS (IN OHMS)
1 1010 991 985 986
2 995 996 1009 994
3 990 1003 1015 1008
4 1015 1020 1009 998
5 1013 1019 1005 993
6 994 1001 994 1005
7 989 992 982 1020
8 1001 986 996 996
9 1006 989 1005 1007
10 992 1007 1006 979
11 996 1006 997 989
12 1019 996 991 1011
13 981 991 989 1003
14 999 993 988 984
15 1013 1002 1005 992
Develop an X chart and an R chart and plot the values. From the charts, what comments
can you make about the process? (Use three-sigma control limits as in Exhibit TN7.7.)
7 In the past, Alpha Corporation has not performed incoming quality control inspections but has
taken the word of its vendors. However, Alpha has been having some unsatisfactory experience
recently with the quality of purchased items and wants to set up sampling plans for the
receiving department to use.
For a particular component, X, Alpha has a lot tolerance percentage defective of 10 percent.
Zenon Corporation, from which Alpha purchases this component, has an acceptable
3. a. p­ = .067.
UCL = .194.
LCL = 0.
b. Stop the process. Something is
wrong. There is wide variation and
two are out of limits.
4. a. Cost to inspect = $.40/unit.
Cost of defect = $.50/unit.
Inspect.
b. Benefit is $.10/unit.
6. X
­­
= 999.1;
UCL = 1014.965;
LCL = 983.235;
R
­
= 21.733;
UCL = 49.551
LCL = 0.
See ISM for graph.
5. Yes, inspecting is
cheaper. Cost to inspect
is $.267/unit. Cost of
defect is $.30/unit.
7. a. 87.1 sample size, round to 88.
b. c = 5.
cha06369_tn07.qxd 2/12/03 7:01 PM Page 318
PROCESS CAPABILITY AND STATISTICAL QUALITY CONTROL technical note 319
quality level in its production facility of 3 percent for component X. Alpha has a consumer's
risk of 10 percent and Zenon has a producer's risk of 5 percent.
a. When a shipment of Product X is received from Zenon Corporation, what sample size
should the receiving department test?
b. What is the allowable number of defects in order to accept the shipment?
8 You are the newly appointed assistant administrator at a local hospital, and your first project
is to investigate the quality of the patient meals put out by the food-service department. You
conducted a 10-day survey by submitting a simple questionnaire to the 400 patients with
each meal, asking that they simply check off that the meal was either satisfactory or unsatisfactory.
For simplicity in this problem, assume that the response was 1,000 returned questionnaires
from the 1,200 meals each day. The results are as follows:
NUMBER OF
UNSATISFACTORY
MEALS SAMPLE SIZE
December 1 74 1,000
December 2 42 1,000
December 3 64 1,000
December 4 80 1,000
December 5 40 1,000
December 6 50 1,000
December 7 65 1,000
December 8 70 1,000
December 9 40 1,000
December 10 75 1,000
600 10,000
a. Construct a p chart based on the questionnaire results, using a confidence interval of
95.5 percent, which is two standard deviations.
b. What comments can you make about the results of the survey?
9 Large-scale integrated (LSI) circuit chips are made in one department of an electronics firm.
These chips are incorporated into analog devices that are then encased in epoxy. The yield is
not particularly good for LSI manufacture, so the AQL specified by that department is 0.15
while the LTPD acceptable by the assembly department is 0.40.
a. Develop a sampling plan.
b. Explain what the sampling plan means; that is, how would you tell someone to do the test?
10 The state and local police departments are trying to analyze crime rates so they can shift their
patrolsfromdecreasing-rateareastoareaswhereratesareincreasing.Thecityandcountyhave
been geographically segmented into areas containing 5,000 residences. The police recognize
that not all crimes and offenses are reported: people do not want to become involved, consider
the offenses too small to report, are too embarrassed to make a police report, or do not take the
time, among other reasons. Every month, because of this, the police are contacting by phone a
random sample of 1,000 of the 5,000 residences for data on crime. (Respondents are guaranteed
anonymity.) Here are the data collected for the past 12 months for one area:
MONTH CRIME INCIDENCE SAMPLE SIZE CRIME RATE
January 7 1,000 0.007
February 9 1,000 0.009
March 7 1,000 0.007
April 7 1,000 0.007
May 7 1,000 0.007
June 9 1,000 0.009
July 7 1,000 0.007
August 10 1,000 0.010
September 8 1,000 0.008
October 11 1,000 0.011
November 10 1,000 0.010
December 8 1,000 0.008
9. a. n = 31.3. (Round sample size to 32.)
b. Random sample 32; reject if more
than 8 are defective.
8. a. p­
= .06;
sp = .0075;
UCL = .075;
LCL = .045.
b. Process is erratic and out of
control.
10. p­
= .0083;
sp = .00287;
UCL = .01396;
LCL = .00270;
Based on p chart, crime rate has
not increased.
cha06369_tn07.qxd 2/12/03 7:01 PM Page 319
320 section 2 PRODUCT DESIGN AND PROCESS SELECTION
Construct a p chart for 95 percent confidence (1.96) and plot each of the months. If the next
three months show crime incidences in this area as
January = 10 (out of 1,000 sampled)
February = 12 (out of 1,000 sampled)
March = 11 (out of 1,000 sampled)
what comments can you make regarding the crime rate?
11 Some citizens complained to city council members that there should be equal protection
under the law against the occurrence of crimes. The citizens argued that this equal protection
should be interpreted as indicating that high-crime areas should have more police protection
than low-crime areas. Therefore, police patrols and other methods for preventing
crime (such as street lighting or cleaning up abandoned areas and buildings) should be used
proportionately to crime occurrence.
In a fashion similar to Problem 10, the city has been broken down into 20 geographic
areas, each containing 5,000 residences. The 1,000 sampled from each area showed the following
incidence of crime during the past month:
AREA NUMBER OF CRIMES SAMPLE SIZE CRIME RATE
1 14 1,000 0.014
2 3 1,000 0.003
3 19 1,000 0.019
4 18 1,000 0.018
5 14 1,000 0.014
6 28 1,000 0.028
7 10 1,000 0.010
8 18 1,000 0.018
9 12 1,000 0.012
10 3 1,000 0.003
11 20 1,000 0.020
12 15 1,000 0.015
13 12 1,000 0.012
14 14 1,000 0.014
15 10 1,000 0.010
16 30 1,000 0.030
17 4 1,000 0.004
18 20 1,000 0.020
19 6 1,000 0.006
20 30 1,000 0.030
300
Suggest a reallocation of crime protection effort, if indicated, based on a p chart analysis.
To be reasonably certain in your recommendation, select a 95 percent confidence level (that
is, Z = 1.96).
12 The following table contains the measurements of the key length dimension from a fuel
injector. These samples of size five were taken at one-hour intervals.
OBSERVATIONS
SAMPLE NUMBER 1 2 3 4 5
1 .486 .499 .493 .511 .481
2 .499 .506 .516 .494 .529
3 .496 .500 .515 .488 .521
4 .495 .506 .483 .487 .489
5 .472 .502 .526 .469 .481
6 .473 .495 .507 .493 .506
7 .495 .512 .490 .471 .504 (continued)
11. p­
= 0.015;
sp = 0.00384;
UCL = 0.0225;
LCL = 0.0075;
Three areas outside UCL warrant
further investigation; four areas
below LCL warrant investigation.
12. X = .499
UCL = .520
LCL = .478
R = .037
UCL = .078
LCL = .000
Process is in control.
cha06369_tn07.qxd 2/12/03 7:01 PM Page 320
PROCESS CAPABILITY AND STATISTICAL QUALITY CONTROL technical note 321
OBSERVATIONS
SAMPLE NUMBER 1 2 3 4 5
8 .525 .501 .498 .474 .485
9 .497 .501 .517 .506 .516
10 .495 .505 .516 .511 .497
11 .495 .482 .468 .492 .492
12 .483 .459 .526 .506 .522
13 .521 .512 .493 .525 .510
14 .487 .521 .507 .501 .500
15 .493 .516 .499 .511 .513
16 .473 .506 .479 .480 .523
17 .477 .485 .513 .484 .496
18 .515 .493 .493 .485 .475
19 .511 .536 .486 .497 .491
20 .509 .490 .470 .504 .512
Construct a three-sigma X chart and R chart (use Exhibit TN7.7) for the length of the fuel
injector. What can you say about this process?
13 C-Spec, Inc., is attempting to determine whether an existing machine is capable of milling
an engine part that has a key specification of 4  .003 inches. After a trial run on this machine,
C-Spec has determined that the machine has a sample mean of 4.001 inches with a
standard deviation of .002 inch.
a. Calculate the Cpk for this machine.
b. Should C-Spec use this machine to produce this part? Why?
A D V A N C E D P R O B L E M
14 Design specifications require that a key dimension on a product measure 100  10 units. A
process being considered for producing this product has a standard deviation of four units.
a. What can you say (quantitatively) regarding the process capability?
b. Suppose the process average shifts to 92. Calculate the new process capability.
c. What can you say about the process after the shift? Approximately what percentage of the
items produced will be defective?
S E L E C T E D B I B L I O G R A P H Y
Alwan, L. Statistical Process Analysis. New York: Irwin/McGraw-Hill,
2000.
Aslup, F., and R. M. Watson. Practical Statistical Process Control: A Tool
for Quality Manufacturing. New York: Van Nostrand Reinhold,
1993.
Feigenbaum,A. V. Total Quality Control. New York: McGraw-Hill, 1991.
Grant, E. L., and R. S. Leavenworth. Statistical Quality Control.
New York: McGraw-Hill, 1996.
Ledolter, J., and C. Burrill. Statistical Quality Control: Strategies and
Tools for Continual Improvement. New York: Wiley, 1999.
Rath & Strong. Rath & Strong's Six Sigma Pocket Guide. Rath & Strong,
Inc., 2000.
Small, B. B. (with committee). Statistical Quality Control Handbook.
Western Electric Co., Inc., 1956.
Taguchi, G. On-Line Quality Control during Production. Tokyo:
Japanese Standards Association, 1987.
Zimmerman, S. M., and M. L. Icenogel. Statistical Quality Control;
Using Excel. Milwaukee, WI: ASQ Quality Press, 1999.
14. a. Cpk =
110 - 100
3 × 4
= .8333
b. Cpk =
92 - 90
3 × 4
= .1667
c. Many defects will be produced.
Left tail is a z -2/4 -.5,
which corresponds to a probability
of 0.1915. Right tail is at z
18/4 4.5, which corresponds to
a probability of approximately .5.
The probability of being inside the
limits is .1915 + .5 .6915, and
the probability of being outside is
1 - .6915 .3085. Approximately
31% of the parts will be defective.
13. a. Cpk = .333
b. No, the machine is
not capable of
producing this part.
F O O T N O T E S
1 D. Birch, "The True Value of 6 Sigma," Quality Progress, April 1993. p. 6.
2 E. L. Grant and R. Leavenworth, Statistical Quality Control (New York: McGraw-Hill, 1964), p. 562.
3 There is some controversy surrounding AQLs. This is based on the argument that specifying some acceptable percentage
of defectives is inconsistent with the philosophical goal of zero defects. In practice, even in the best QC companies, there
is an acceptable quality level. The difference is that it may be stated in parts per million rather than in parts per hundred.
This is the case in Motorola's six-sigma quality standard, which holds that no more than 3.4 defects per million parts are
acceptable.
4 See, for example, H. F. Dodge and H. G. Romig, Sampling Inspection Tables--Single and Double Sampling (New York: John
Wiley & Sons, 1959); and Military Standard Sampling Procedures and Tables for Inspection by Attributes (MIL-STD-105D)
(Washington, DC: U.S. Government Printing Office, 1983).
cha06369_tn07.qxd 2/12/03 7:01 PM Page 321